The MPEG-4 standard is one of the most popular video formats and it defines an audiovisual scene as a coded representation of audiovisual objects (A/V) that have certain relationships in space and time. Coding of arbitrarily-shaped video objects is one of the most important characteristics of the MPEG-4 standard. Each frame of an MPEG-4 sequence is composed of a number of objects, where each object is usually of arbitrary shape. However, it is different from other standards such as MPEG-1, 2 and H.26x in that each arbitrarily-shaped object is coded independently of other objects by the sophisticated MPEG-4 encoder. In this thesis, we will present improvements in two different sub-areas of arbitrary shape video coding, namely motion coding and texture coding....[ Read more ]

The MPEG-4 standard is one of the most popular video formats and it defines an audiovisual scene as a coded representation of audiovisual objects (A/V) that have certain relationships in space and time. Coding of arbitrarily-shaped video objects is one of the most important characteristics of the MPEG-4 standard. Each frame of an MPEG-4 sequence is composed of a number of objects, where each object is usually of arbitrary shape. However, it is different from other standards such as MPEG-1, 2 and H.26x in that each arbitrarily-shaped object is coded independently of other objects by the sophisticated MPEG-4 encoder. In this thesis, we will present improvements in two different sub-areas of arbitrary shape video coding, namely motion coding and texture coding.

Motion coding is a crucial process to compress digital video. In order to eliminate the temporal redundancy among successive video frames, the MPEG-4 standard recommends the full search (FS) block-matching algorithm to evaluate exhaustively all possible candidate macroblocks to get the motion vector for the motion-compensated interframe coding. However, FS may use significant percentage of the computing power of the encoder. In this paper, we propose several novel improvements for fast motion estimation algorithm that is especially tailored for arbitrarily shaped objects. How to solve the problem of content-based or segmented-based coding is another essential issue of our thesis. Since arbitrary shape video object is decomposed into a group of macroblocks, some macroblocks at the border of video object (so-called boundary macroblock) may cause the problem of DCT computation. Therefore, a special padding technique called SmartPad is extended in this thesis to solve the problem of interframe texture coding.

Finally, we combine our two contributions to form a novel arbitrarily-shaped video coding. Complete and extensive simulations are provided in this thesis to demonstrate the performance.